The present invention relates to a bi-directional multi-drop bus system, and a memory module and a memory system using the bi-directional multi-drop bus system.
As the semiconductor fabrication method and the circuit design technique have improved, the bandwidth of the dynamic random access memory (DRAM) has been considerably increased and the data transmission method also has been continuously improved. The increase of the bandwidth influences the direct effect to the operation speed of the input/output (I/O) interface circuit. The operation speed of the I/O interface circuit is determined by the signal integrity of the data channel and the timing margin of the I/O interface circuit.
On the basis of the transmission speed per pin, the bandwidth of memory systems such as a synchronous dynamic random access memory (SDRAM) and double data rate (DDR) 1/2/3 SDRAMs having a multi-drop bus system have been developed in order of 166 Mbps, 333 Mbps, 400 Mbps, 800 Mbps, and 1600 Mbps. The memory systems of SDRM and DDR 1/2/3 SDRAMs have been employed the stub series terminated logic (SSTL) for improving the signal integrity. The SSTL method is the I/O signaling method including Z0/2 stubs resistor. The conventional memory systems employing the SSTL have the problem of the reflected wave induced by the impedance mismatch in the stub. The reflected wave causes the inter symbol interference (ISI). The reflected wave causing the ISI can seriously damage the signal integrity and, thus, it is impossible to implement the memory systems of the large bandwidth, e.g., greater than 2 Gbps.
FIGS. 1A to 1D are diagrams for explaining the concept of the SSTL employed by the conventional memory systems having a multi-drop bus system.
As shown in FIG. 1A, a first signal SIG1 from a memory controller not shown, is transmitted to a first stub 101. As shown in FIG. 1B, a first element SIG11 of the first signal SIG1 is transmitted to a first dual inline memory module (DIMM) 111 and a second element SIG13 of the first signal SIG1 is transmitted to a second DIMM 113. The rest of the first signal SIG1 is re-transmitted to the memory controller as a first reflected wave REF1 caused by the impedance mismatch. Next, as shown in FIG. 1C, a part of the first element SIG11 transmitted to the first DIMM 111 is reflected and re-transmitted to the memory controller as a second reflected wave REF11. In the similar way, a part of the second element SIG13 transmitted to the second DIMM 113 is reflected and re-transmitted to the memory controller as a third reflected wave REF13. Herein, when a second signal SIG2 is transmitted from the memory controller, the second and the third reflected waves REF11 and REF13 of the first signal SIG1 interfere with the transmission of the second signal SIG2. Referring to FIG. 1D, a first element SIG21 of the second signal SIG2 is transmitted to the first DIMM 111 and a second element SIG23 of the second signal SIG2 is transmitted to the second DIMM 113. The rest of the second signal SIG2 is re-transmitted to the memory controller as a fourth reflected wave REF2. However, the second and the third reflected waves REF11 and REF13 of the first signal SIG1 (denoted with the reference number 131 in FIG. 1D) interfere with the transmission of the second signal SIG2 and the ISI, accordingly, occurs. For this reason, the data transmission cannot be performed reliably for the memory systems employing the SSTL. Therefore, the conventional SSTL may not be applicable for next generation memory systems which are expected for large bandwidth requirement, e.g., 3.2 Gbps or 6.4 Gbps.
In order to solve the abovementioned problems, a fully buffered DIMM (FB-DIMM) technique is introduced. The FB-DIMM defined by Joint Electron Device Engineering Council (JEDEC) is the memory module embedded with a special memory controller referred to an advanced memory buffer (AMB). The FB-DIMM technique employs a point-to-point signaling method of a daisy chain memory structure to solve the problems induced by the reflected waves of the SSTL. The FB-DIMM technique includes the AMB between the memory controller and the memory module. Therefore, the data transmission between the memory controller and the memory module is performed via the AMB. The FB-DIMM technique has a serial interface between the AMB and the memory controller. The AMB compensates for signal deterioration by buffering and resending the signal. In addition, the AMB also offers an error correction, without posing any overhead on the processor or the memory controller. By using the serial interface, it is possible for the FB-DIMM technique to solve the problems caused by the reflected waves. However, the FB-DIMM technique introduces the latency to the memory request and requires additional power consumption for the buffering operation.
Another memory system referred to as an extreme data rate (XDR) memory system is introduced by Rambus Inc. The XDR memory system employs a dynamic point-to-point (DPP) signaling method to maintain the signal integrity and solve the problems caused by the serial interface.
FIGS. 2A and 2B are block diagrams illustrating an operation of the XDR memory system.
A basic XDR memory system shown in FIG. 2A includes a memory controller 501, e.g., an application-specific integrated circuit (ASIC), a 32-bit DIMM 503, and a continuity module 505. The continuity module 505 provides an electric connection for DPP signaling method. In case the XDR memory system transmits 4-bit data, the 32-bit DIMM 503 directly transmits 2-bit data to the memory controller 501 and transmits rest 2-bit data to the memory controller 501 via the continuity module 505. Meanwhile, the continuity module 505 can be replaced with a 32 bit DIMM 507 as shown in FIG. 2B. In this case, each of the 32-bit DIMMs 503 and 507 directly transmits 2-bit data to the memory controller 501 when the XDR memory system transmits 4-bit data. As mentioned above, the XDR memory system employs the DPP signaling method to maintain the signal integrity and solve the problems caused by the serial interface. However, the bandwidth of each memory module of the XDR memory system using the DPP signaling method is decreased as the number of the module increases.
A method to solve the above problem has been suggested before: for example, J. Dally and J. W. Poulton “Digital Systems Engineering”, p. 106˜109, Cambridge Univ. Press, Jun. 28, 1998.
FIG. 3 is a schematic diagram describing the suggested method, which removes the reflect wave from the entire channel by connecting resistor network to the stub of the multi-drop channel.
Referring to FIG. 3, the suggested method matches the impedance of every transmission line coupled to each of stubs. By matching impedance of every transmission line, the reflected wave may not occur regardless of which one of transmission lines is selected to transmit the signal. Accordingly, the suggested method improves the signal integrity. However, the voltage amount provided to each memory module is different according to the location of each memory module in the suggested method. For the write operation, the memory module furthest from the memory controller receives the smallest amount of voltage. For the read operation, the data transmitted from the furthest memory module has the smallest voltage level. In order to compensate the voltage difference and provide equivalent amount of voltage to each memory module, the circuitry for each memory module is required to be designed differently from one another. For this reason, it is difficult to apply the suggested method to the memory interface channel.